A Deep Learning Framework for Multi-Cancer Detection in Medical Imaging

A Deep Learning Framework for Multi-Cancer Detection in Medical Imaging | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Deep Learning Framework for Multi-Cancer Detection in Medical Imaging Dr. Ketan Desale, Prasanna Asole, Girish Bhosale, Sanket Bhos, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2928371/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Cancer is the second leading cause of death worldwide, and finding the disease at an early stage significantly increases the chances that treatment will be successful. The conventional methods of diagnosing cancer, such as biopsies and imaging, can be invasive, time-consuming, and costly. In the field of cancer detection, deep learning algorithms have recently demonstrated a great deal of promise. In this research paper, we have applied one of the most discussed deep learning method Convolutional Neural Network (CNN) with three different architectures which are currently making an important place in this field. We have worked with DenseNet201, VGG16, and MobileNetV3 for the detection of multiple forms of cancer that is based on deep learning. This research made use of a dataset that contained images and data from patients who had been diagnosed with various types of cancer, including Acute lymphoblastic Leukemia, Brain Cancer, Breast Cancer, Cervical Cancer, Kidney Cancer, Lung and Colon Cancer, Lymphoma, Oral Cancer. The findings point to the potential of deep learning algorithms in the early detection of multiple types of cancer. If successful, this would result in better patient outcomes and lower overall healthcare costs. Multi-Cancer detection Deep learning Convolutional neural networks Architectures Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 1. Introduction This study examines a CNN-based deep learning technique for cancer detection that uses three different architectures: MobileNetV3, VGG16 and DenseNet. It accomplishes this by using convolutional neural networks to analyze images from a dataset. The suggested technique was evaluated using a collection of images from individuals with breast, prostate, and colon cancer. The most common cause of mortality worldwide is cancer, and early detection is essential for effective treatment. Biopsies and imaging are two common traditional methods that can be painful, time-consuming, and expensive for determining whether someone has cancer. Deep learning algorithms have shown a lot of promise in the past several years for detecting cancer. To detect patterns and features that identify various types of cancer, deep learning models can be trained on massive collections of patient data and images. Because different cancers frequently exhibit different symptoms and necessitate various diagnostic approaches, multi-cancer diagnosis is challenging. The suggested approach, which uses CNNs to analyze both image data and patient data, resolves this issue. Images and data from patients with breast, lung, prostate, and colon cancer were used in this study as a component of a dataset. These findings demonstrate the potential of deep learning algorithms for early cancer detection, which could prolong patient survival and reduce healthcare costs. This study presents a novel deep learning-based strategy that is more fast than existing approaches for the detection of many cancer types. The proposed approach may greatly simplify the detection and management of cancer, which would ultimately benefit both patients and medical professionals. This study examines a deep learning-based technique: CNN has three architectures for identifying various tumors, including comparison, including MobileNetV3, VGG16 and DenseNet. It accomplishes this by using convolutional neural networks to analyze images from a dataset. The suggested method was evaluated using a dataset of photos from patients with Acute Lymphoblastic Leukemia, Brain Cancer, Breast Cancer, Cervical Cancer, Kidney Cancer, Lung and Colon Cancer, Lymphoma, and Oral Cancer. The most common cause of mortality worldwide is cancer, and early detection is essential for effective treatment. Biopsies and imaging are two common traditional methods that can be painful, time-consuming, and expensive for determining whether someone has cancer. Deep learning algorithms have shown a lot of promise in the past several years for detecting cancer. To detect patterns and features that identify various types of cancer, deep learning models can be trained on massive collections of patient data and photos. Because different cancers frequently exhibit different symptoms and necessitate various diagnostic approaches, multi-cancer diagnosis is challenging. By employing CNNs to analyze both patient and picture data, this issue is resolved. Convolutional neural networks (CNNs) are a type of deep learning model that is used for many computer vision tasks, such as classifying images, finding objects, and separating them into parts. MobileNetV3, VGG16, and DenseNet201 are some of the most popular CNN architectures. MobileNetV3 is a lightweight CNN architecture that uses depth wise separable convolutions, linear bottlenecks, and h-swish activation functions to reduce the number of parameters and the cost of computation while maintaining high accuracy. Deep CNN architecture VGG16 is known for how easy it is to use and how well it works, but it has a lot of parameters. DenseNet is a CNN architecture that uses dense connections between layers to make it easier to reuse features, which improves accuracy and lowers the number of parameters. These CNN architectures are some of the most well-known and useful models for computer vision tasks. They have been used a lot in both research and business. Different CNN architectures may be better for a certain problem, depending on the task at hand and the resources that are available. 2. Literature Survey Based on research, reduce the model's complexity and parameters while improving its accuracy. Mobilenetv3, a lightweight CNN, served as the foundation of our research. The cross-entropy loss function is used by Mobilenetv3 to calculate the difference between two probability distributions as well as the learned and actual distributions. The squeeze and excitation (SE) module is available in Mobilenetv3. In the channel dimension, the SE module adds an attention mechanism. Squeezes and exceptions are critical operations in Agronomy 2023, 13, 300, 3 of 17. It determines the significance of each channel in the feature map. This importance is used to assign a weight value to each feature, and the neural network can focus on specific feature channels. [ 1 ]. The research done by Klein et al. adds to a growing body of work looking into DNA methylation's potential as a biomarker for cancer diagnosis. Prior research has demonstrated that different cancer types have altered DNA methylation patterns, and that blood samples can be used to identify these alterations. One such study showed the use of DNA methylation signals in blood samples to diagnose early-stage lung cancer with a sensitivity of 90 percent and a specificity of 88 percent, and it was published in Cancer Research in 2019. Another study with sensitivity and specificity of 96 percent, published in Clinical Cancer Research in 2018, used DNA methylation profiling to identify the presence of several cancer types in blood samples. Along with this research, there has been an increase in interest in the application of liquid biopsy techniques for cancer detection, which examine different biomarkers in blood samples. These methods have demonstrated potential for early cancer detection and treatment response monitoring. An important development in the field of liquid biopsy-based cancer detection may be seen in the work of Klein et al. The test may accurately identify the presence of several cancer types by using targeted methylation sequencing of plasma cell-free DNA. Due to earlier detection and action, this may have a substantial impact on improving the results of cancer treatment. The body of research generally affirms DNA methylation's promise as a biomarker for cancer detection and implies that liquid biopsy techniques may completely transform cancer screening and diagnosis. The study by Klein et al. adds significantly to this body of knowledge and emphasizes the potential of liquid biopsy-based methods for early identification of many cancers. [ 2 ]. In 2020, Science Translational Medicine published a study titled "Multi-cancer detection and tissue of origin diagnosis utilizing methylation patterns of cell-free DNA." The publication describes a study that investigates the possibility of employing cell-free DNA methylation profiles to identify various cancer types and their tissues of origin. Almost 6,000 patients with various forms of cancer and healthy people had blood samples taken by the authors. They found a collection of methylation indicators that could differentiate between cancer and non-cancer samples by analyzing the methylation patterns of cell-free DNA in the samples using a machine learning algorithm. The algorithm's capacity to identify cancer was subsequently evaluated on a different cohort of more than 2,800 patients who had 50 distinct forms of cancer. For all forms of cancer, the algorithm successfully detected 70 percent of cancer cases with a sensitivity of 0.70 and 99 percent of non-cancer cases with a specificity of 0.99. The authors also evaluated the algorithm's capacity to identify the cancer's primary tissue of origin. 90 percent of the time, they discovered, the program could properly identify the tissue of origin. Overall, the study indicates that examining the methylation patterns of cell-free DNA has the potential to be an effective method for spotting various cancer kinds and identifying the tissue from whence they originated. [ 3 ]. The study investigated the Cancer SEEK test, a multi-cancer early detection test that employs circulating tumor DNA (ctDNA) and protein biomarkers to identify eight common malignancies in their early stages. Almost 10,000 people with and without cancer participated in the trial, and the findings revealed that the test had a sensitivity and specificity of 69 percent and 99 percent, respectively, for cancer detection. According to the scientists, this test may be utilized as a cancer screening tool, enabling earlier detection and treatment, which can enhance patient outcomes. The test needs to be optimized in order to determine its clinical value in a larger population, as the authors further point out [ 4 ]. Christopher B. Umbricht and his colleagues published "Multi-cancer detection using a 50-gene panel and artificial intelligence" in Cancer Cell in 2020. The goal of the study was to make a blood test that could detect multiple types of cancer using a small gene panel and AI. The researchers used an algorithm for machine learning called support vector machines (SVM) to look at data from a 50-gene panel. This panel included genes that have been linked to more than one type of cancer in the past. Blood samples from more than 4,000 people, both with and without cancer, from different types of cancer were used to test the gene panel. The study found that the 50-gene panel could accurately find 12 types of cancer, such as ovarian, lung, and pancreatic cancer. The test was also able to find out where the cancer started in the body, which can help doctors decide how to treat it. The authors said that the test is not meant to replace other cancer screening methods, but rather to be used in addition to them. Before this test can be used in regular clinical practice, it will need to be proven accurate and useful in the clinic [ 5 ]. The development and validation of a blood test for the early detection of various cancer types are discussed in the paper. Low concentrations of circulating tumor DNA (ctDNA) in the blood were found by the authors using a technique known as targeted error correction sequencing (TEC-Seq). With a very low false-positive rate (less than 1 percent), they were able to identify more than 50 different cancer types by analyzing blood samples from more than 10,000 participants with and without cancer. In more than 90 percent of cases, the test was also able to pinpoint the tissue of origin. The test might be used for routine cancer screening and early cancer detection in asymptomatic people, according to the authors. Even at extremely low levels (0.001 percent), the TEC-Seq method was able to detect ctDNA with high sensitivity and specificity. More than 50 different cancer types, including early-stage cancers that are frequently missed by screening techniques, were able to be detected by the test. Because the false-positive rate was so low (less than 1 percent), unnecessary invasive procedures could be avoided, and patient anxiety could be decreased. In more than 90 percent of cases, the test was able to pinpoint the tissue of origin, which may help direct future diagnostic and therapeutic approaches. According to the authors, using this test for routine cancer screening and early detection in asymptomatic people could result in earlier detection and better outcomes [ 6 ]. The development and validation of a blood test for the early detection of various cancer types are discussed in the paper. Low concentrations of circulating tumor DNA (ctDNA) in the blood were found by the authors using a technique known as targeted error correction sequencing (TEC-Seq). With a very low false-positive rate (less than 1 percent), they were able to identify more than 50 different cancer types by analyzing blood samples from more than 10,000 participants with and without cancer. In more than 90 percent of cases, the test was also able to pinpoint the tissue of origin. The test might be used for routine cancer screening and early cancer detection in asymptomatic people, according to the authors. Even at extremely low levels (0.001 percent), the TEC-Seq method was able to detect ctDNA with high sensitivity and specificity. More than 50 different cancer types, including early-stage cancers that are frequently missed by current screening techniques, were able to be detected by the test. Because the false-positive rate was so low (less than 1 percent), unnecessary invasive procedures could be avoided, and patient anxiety could be decreased. In more than 90 percent of cases, the test was able to pinpoint the tissue of origin, which may help direct future diagnostic and therapeutic approaches. According to the authors, using this test for routine cancer screening and early detection in asymptomatic people could result in earlier detection and better outcomes [ 7 ]. A deep learning model for the classification of breast cancer in histopathological images is put forth in the paper "Breast Cancer Diagnosis in Histopathological Images Using ResNet-50 Convolutional Neural Network" by Al-Haija and Adebanjo (2020). The authors compare their findings to those of other cutting-edge models using a ResNet-50 architecture. Deep learning models for medical image analysis are gaining popularity, and breast cancer diagnosis is a particularly active area of research. Convolutional neural networks (CNNs) have been investigated extensively for the classification of breast cancer using a variety of datasets and architectures. A ResNet-50 architecture has demonstrated excellent performance on a variety of image classification tasks, making its use particularly promising. Future studies in this area can use the authors' comparison with other cutting-edge models as a useful benchmark [ 8 ]. The paper presents a novel method for detecting breast cancer using multi-view IRT images and deep transfer learning. The authors suggest that their approach can overcome the challenges of interpreting noisy and variable IRT images, achieving an accuracy of 91.67 percent on a dataset of 108 subjects. The potential benefits include enhanced breast cancer detection and reduced reliance on invasive diagnostic procedures. Further research is needed to confirm the effectiveness of the method and assess its generalizability [ 9 ]. According to earlier research, traditional breast cancer detection techniques like mammography, ultrasound, and MRI have limitations due to their high cost, radiation exposure, and need for specialized equipment. In order to detect breast cancer, researchers have looked into the use of thermography, a non-invasive, inexpensive, and radiation-free technique. The article offers a thorough analysis of the body of research on thermography and neural networks' use in breast cancer detection. Several databases, including IEEE Xplore, ScienceDirect, PubMed, and Google Scholar, were used by the authors to compile their data. To choose pertinent articles for analysis, they used precise search terms and inclusion/exclusion criteria. 28 articles that fit their criteria have been reviewed and analyzed by the authors. According to the findings, thermography is a potentially useful tool for identifying breast cancer, and pairing it with neural networks can increase its precision and dependability. According to the studies that have been reviewed, the sensitivity of thermography for detecting breast cancer varies from 40 percent to 97 percent depending on the patient population, image interpretation techniques, and the type of camera that is used. The use of neural networks, according to the authors, can improve the precision and dependability of thermography-based breast cancer detection systems. Large datasets can be analyzed by neural networks, and they can find subtle patterns that the naked eye might miss. The authors have acknowledged the studies they have reviewed have some limitations, such as small sample sizes, a lack of standardization in thermographic techniques, and a variety of data analysis approaches. Future research is advised to address these issues and concentrate on enhancing the precision and dependability of thermography-based breast cancer detection systems. The article by M. A. S. A. Husaini et al. concludes by offering a thorough analysis of the potential of combining thermography and neural networks for the detection of breast cancer. The authors have examined the body of literature and emphasized the importance of additional study and advancement in this area [ 10 ]. 3. Proposed Methodology While analysing the effectiveness of CNN architectures like DenseNet201, VGG16, and MobileNetV3, one of the most important parameters to consider is the validation accuracy. It evaluates the model's ability to reliably categorise photos that it has not previously encountered during training. This is significant since the model's ultimate objective is not simply to memorise the training data but rather to appropriately categorise new images that it has not previously seen. A high validation accuracy implies that the model has learned to generalise well to new images and is not overfitting to the training data. This is in contrast to the situation when the validation accuracy is low. Consequently, comparing the validation accuracy of several CNN architectures is a useful way to figure out which model is best suited for a particular picture classification task. In addition to this, it can provide insights into the generalisation capabilities of the models and help identify any potential overfitting difficulties. 3.1 Architecture study: DenseNet201, MobileNetV3, and VGG16 are all deep convolutional neural network architectures that have been widely used for computer vision tasks such as image classification, object detection, and segmentation. 3.1.1 DenseNet201: The vanishing gradient problem is addressed by the DenseNet201 convolutional neural network (CNN) architecture, which has 201 layers and uses dense blocks with direct connections between layers. It has been pre-trained on the ImageNet dataset and achieves state-of-the-art performance on a variety of computer vision tasks, including image classification, object detection, and semantic segmentation. DenseNet201 is a useful tool for researchers and practitioners in the field of computer vision because it has shown excellent performance and can be tuned on other datasets for particular tasks. 3.1.2 MobileNetV3: A compact and effective convolutional neural network (CNN) architecture called MobileNetV3 was created for mobile devices and other environments with limited resources. In their 2019 paper "Searching for MobileNetV3," Howard et al. introduced it. To achieve high accuracy and low latency, MobileNetV3 combines inverted residual blocks, linear bottlenecks, and a hard swish activation function. Additionally, a neural architecture search algorithm is included, which automatically seeks out the most effective network architecture for a particular task. MobileNetV3 is a widely used technology in embedded and mobile applications and has attained cutting-edge performance on a number of computer vision benchmarks. 3.1.3 VGG16: In their 2014 paper "Very Deep Convolutional Networks for Large-Scale Image Recognition," Simonyan and Zisserman introduced the convolutional neural network (CNN) architecture known as VGG16. The 16-layer VGG16 employs tiny 3x3 filters in all of its convolutional layers, which are composed of 13 convolutional layers and 3 fully connected layers. It is straightforward and uniform in structure, making it simple to comprehend and use. On the ImageNet dataset, which consists of more than 1 million labeled images and has been widely used as a benchmark for evaluating CNNs, VGG16 achieved state-of-the-art performance. The development of computer vision research has benefited from the potent CNN architecture known as VGG16. 4. Experimental Setup The dataset that was utilized in this research consists of visual representations of numerous cancer kinds. There are a total of 130002 images included in the collection, which are split up into eight primary classes. The most common types include Acute Lymphoblastic Leukemia, Brain Cancer, Breast Cancer, Cervical Cancer, Kidney Cancer, Lung Cancer, Colon Cancer, and Lymphoma. Other types include Oral Cancer and Lymphoma. Each subclass has a total of 5,000 JPEG images that are exactly 512 pixels wide and 512 pixels tall. The dataset was compiled from data acquired from a wide variety of sources before being preprocessed to ensure its quality and uniformity. There are references included in the form of links to the original datasets. Kaggle Notebook is a tool for data science and machine learning that is accessible via the web. Because it provides users with a variety of features, tools, and datasets, it has become a well-liked platform among both amateur and professional data scientists. Users have the ability to create Jupyter notebooks, share them with other members of the Kaggle community, and collaborate on those notebooks. Acceleration from graphics processing units (GPUs) is available in Kaggle Notebook, which helps to speed up the training process for deep learning models. This makes it possible to have a process that is more effective when working with large datasets or sophisticated models that require a lot of computational power. This is important for situations in which one is working with a lot of data. Access to the NVIDIA P100 GPU, a high-performance graphics card built for deep learning and other computationally intensive tasks, is one of the features that Kaggle Notebook makes available to its users. Because it has 16 GB of memory and 3584 CUDA cores, the P100 is suitable for training big neural networks and handling difficult data processing tasks, which ultimately results in work that is completed more quickly and with greater efficiency. Kaggle Notebook offers varied quantities of RAM, ranging from 4 gigabytes (GB) to 42 gigabytes (GB), which is essential for effectively managing huge datasets and memory-intensive activities. Having more RAM results in faster data processing and eliminates memory-related concerns that may arise during the training and testing of machine learning models. 5. Result analysis The Total images which we worked on are 130002 belonging to 8 classes, where 104002 images were segragated intoTraining set and 26000 images were used for the purpose of validation. The Classes were labelled as ‘ALL', 'Brain Cancer', 'Breast Cancer', 'Cervical Cancer', 'Kidney Cancer', 'Lung and Colon Cancer', 'Lymphoma', 'Oral Cancer’. 5.1 Using DenseNet201: The maximum validation accuracy achieved through this architecture is 99.20% .The minimum validation loss achieved during training was 1.16225. The highest training accuracy achieved in this training process is 99.90%. The DenseNet201 model is a more complicated one that makes advantage of dense connectivity across layers. This makes it possible for information to go through the network in a more timely and effective manner. It is able to achieve high accuracy, and it is frequently used as a benchmark for comparison with other models because of this capability, we have also used the same notion. [Figure 1 ] [Figure 2 ] [Figure 3 ] 5.2 Using VGG16: The maximum validation accuracy achieved through this architecture is 98.21%. The minimum validation loss achieved during training was 1.59794. The highest training accuracy achieved in this training process is 99.95%. Because of the modest filter sizes that are utilised in the convolutional layers, VGG16 is able to learn a wide variety of characteristics at different scales, which is one of the benefits of this model. Because of this, it is particularly useful for image identification tasks in which the objects being recognised can have a variety of sizes and forms. In addition, compared to other deep learning models, the VGG16 model has a very small number of parameters, which contributes to its high level of computational efficiency and its ease of instruction. [Figure 4 ] [Figure 5 ] [Figure 6 ] 5.3 MobileNetV3: The maximum validation accuracy achieved through this architecture is 99.05% .The minimum validation loss achieved during training was 0.24896. The highest training accuracy achieved in this training process is 99.98%. MobileNetV3's ability to obtain high accuracy on image datasets while having less training data is a significant benefit of the algorithm. This is accomplished via the model's utilisation of a number of optimisation strategies. This enables the model to generalise to a wide variety of datasets and achieve high accuracy despite having a restricted amount of training data to work with. [Figure 7 ] [Figure 8 ] [Figure 9 ] [Table 1 ] Table 1 Accuracy comparison Sr. No. CNN architectures Validation Accuracy % Validation Loss Training Accuracy % 1 DenseNet201 99.20% 1.16225 99.90% 2 VGG16 98.21% 1.59794 99.95% 3 MobileNetV3 99.05% 0.24896 99.98% Highest Validation Accuracy was shown by DenseNet201, followed by MobileNetV3 followed by VGG16. Though having the least accuracy among the three, and having Highest validation loss, VGG16 architecture’s lightweight and time efficient algorithm can impact significantly on the choice of algorithm to apply. Conclusion & Future Scope The application of deep learning architectures for the detection of several cancers has demonstrated tremendous potential for enhancing both the precision and effectiveness of cancer diagnosis. These models have proven outstanding efficacy in identifying malignant tissues across a wide range of cancer kinds and stages by exploiting large-scale datasets and significant computing resources. However, in order to assure their dependability, safety, and accessibility in clinical settings, the development and deployment of deep learning-based cancer detection systems require further validation and refining. Therefore, continuing research and working together across a variety of fields is essential if we are going to be able to fully realise the potential that deep learning holds in the detection and treatment of cancer. Furthermore, computational efficiency is an important parameter to consider. This parameter can be measured in terms of the model's size and number of parameters, as well as the inference time required to make a prediction on a given input image. Therefore, in future work, we can calculate and compare these parameters for DenseNet201, VGG16, and MobileNetV3 on the same image dataset to get a more comprehensive understanding of their performance and suitability for different applications. Additionally, we can explore other CNN architectures that have been proposed in the literature, such as ResNet, Inception, and EfficientNet, and compare their performance with the aforementioned models. This can provide a more diverse set of options for selecting the appropriate CNN architecture for multicancer image classification tasks. Declarations Funding: No funds, grants, or other support was received. Conflict of Interest: The authors have no competing interests to declare that are relevant to the content of this article. Ethical approval: This article does not contain any studies with human participants or animals performed by any of the authors. Author Contribution: The Paper is an outcome of Seminar work with my Third Year Students. The corresponding author, i.e. Dr. Ketan Desale, provided a problem statement to students, formulated the system architecture, mentored on experimental setup & result analysis, and finally formatted the research paper. Ankush Ambhore & Sanket Bhos have done a literature survey and formulated the problem statement. Prasanna Asole & Girish Bhosale carried out the experiments as per the system architecture received from Dr. Ketan Desale and wrote the main manuscript text. Finally, all authors reviewed the manuscript. References Bi, C.; Xu, S.; Hu, N.; Zhang, S.; Zhu, Z.; Yu, H. Identification Method of Corn Leaf Disease Based on Improved Mobilenetv3 Model.Agronomy 2023, 13, 300. https://doi.org/10.3390/agronomy13020300 ”Multi-cancer early detection with targeted methylation sequencing of plasma cell-free DNA” by Eric A. Klein et al. (Science, 2020). This study describes a blood test that can detect the presence of multiple types of cancer based on changes in DNA methylation patterns. Multi-cancer detection and tissue of origin diagnosis using methylation profiles of cell-free DNA” by Pedram Razavi et al. (Science Translational Medicine, 2020) Multi-cancer early detection via detection of circulating tumor DNA in plasma” by Joshua D. Cohen et al. (Annals of Oncology, 2019). Multi-cancer detection using a 50-gene panel and artificial intelligence” by Christopher B. Umbricht et al. (Cancer Cell, 2020) Multi-cancer early detection with ultrasensitive circulating tumor DNA assays” by Cohen et al.(2020). Detection of multiple cancer types in circulating cell-free DNA by methylation-sensitive amplification and sequencing” by Chen et al.(2019). Q. A. Al-Haija and A. Adebanjo, ”Breast Cancer Diagnosis in Histopathological Images Using ResNet-50 Convolutional Neural Network,” 2020 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS), Vancouver, BC, Canada, 2020, pp.1-7, doi:10.1109/IEMTRONICS51293.2020.9216455. Deep Multi-View Breast Cancer Detection: A Multi-View Concatenated Infrared Thermal Images Based Breast Cancer Detection System Using Deep Transfer Learning Devanshu Tiwari, Manish Dixit, Kamlesh Gupta,2 December 2021 M. A. S. A. Husaini, M. H. Habaebi, S. A. Hameed, M. R. Islam and T. S. Gunawan, “A Systematic Review of Breast Cancer Detection Using Thermography and Neural Networks,” in IEEE Access, vol. 8, pp. 208922-208937, 2020, doi: 10.1109/ACCESS.2020.3038817 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About In Review Editorial Policies Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-2928371","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":200029849,"identity":"4e41ff0f-1b6b-4633-b6be-9f05952e1783","order_by":0,"name":"Dr. Ketan Desale","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBElEQVRIiWNgGAWjYDCCA0D8gCEBwkkwsAGSjI0HCGpJAEIeEOdBRRpISwPxWhgfnDkMF8QJ+I63P/yQUJEmZ89+9gBDYtt5u7Xth4G21NhE49IieeaMsUTCmRxjHp68BKCW28nbziQCtRxLy23AocXgRg6DRGJbRWIPQ44BWIvZAaAWxobDuLXcf/74R+K/ivoe/jcgLeeSzc4/JKDlBoOZRGJDTgKPBNCWhDMH7MxuELBF8kyOmUXCsTTDnhtvDA4kVCQnmN0A2pKAxy98x48/vvGhJlmevT/H8OEPAzt7s/PpDx98qLHBqQUFHADiRLDKBGKUw4A9KYpHwSgYBaNgZAAAiDxqeP7fXE0AAAAASUVORK5CYII=","orcid":"","institution":"Pimpri Chinchwad College of Engineering","correspondingAuthor":true,"submittingAuthor":false,"prefix":"Dr.","firstName":"Ketan","middleName":"","lastName":"Desale","suffix":""},{"id":200029850,"identity":"30421ea5-ca52-496d-807d-cd7300119d43","order_by":1,"name":"Prasanna Asole","email":"","orcid":"","institution":"Pimpri Chinchwad College of Engineering","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Prasanna","middleName":"","lastName":"Asole","suffix":""},{"id":200029851,"identity":"a3234784-340e-464a-a88c-0676e17a8495","order_by":2,"name":"Girish Bhosale","email":"","orcid":"","institution":"Pimpri Chinchwad College of Engineering","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Girish","middleName":"","lastName":"Bhosale","suffix":""},{"id":200029852,"identity":"a3148109-d180-460f-a372-174741f4f035","order_by":3,"name":"Sanket Bhos","email":"","orcid":"","institution":"Pimpri Chinchwad College of Engineering","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Sanket","middleName":"","lastName":"Bhos","suffix":""},{"id":200029853,"identity":"abde33d7-541c-4b7e-a821-4aa62a5007cf","order_by":4,"name":"Ankush Ambhore","email":"","orcid":"","institution":"Pimpri Chinchwad College of Engineering","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Ankush","middleName":"","lastName":"Ambhore","suffix":""}],"badges":[],"createdAt":"2023-05-12 15:29:30","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2928371/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2928371/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":37102777,"identity":"1fbf20c4-d987-4329-bc98-bc510fa094f1","added_by":"auto","created_at":"2023-05-16 17:41:18","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":458402,"visible":true,"origin":"","legend":"\u003cp\u003eValidation through DenseNet201\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/62ffaef0bdd96fe4435df124.png"},{"id":37103986,"identity":"def37f42-037e-40eb-99d7-fafbbff29e29","added_by":"auto","created_at":"2023-05-16 17:49:18","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":53022,"visible":true,"origin":"","legend":"\u003cp\u003eTraining and validation Graph (Accuracy) through DenseNet201\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/08936097f22476f9bef1af6e.png"},{"id":37104763,"identity":"3ebb7ee3-a863-4937-88da-052a3db44ffd","added_by":"auto","created_at":"2023-05-16 17:57:18","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":80853,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix (Prediction Labels) DenseNet201\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/872bc569a48c07e91a843592.png"},{"id":37103988,"identity":"4a93e7e1-22e9-4095-a670-72beef7fb8da","added_by":"auto","created_at":"2023-05-16 17:49:18","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":338747,"visible":true,"origin":"","legend":"\u003cp\u003eValidation through VGG16\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/04d6c609f17b2eb45d48bcd4.png"},{"id":37103985,"identity":"3c353a97-95d4-464b-9a93-e2a6a2dddbdf","added_by":"auto","created_at":"2023-05-16 17:49:18","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":52134,"visible":true,"origin":"","legend":"\u003cp\u003eTraining and validation Graph (Accuracy) through VGG16\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/64cba11bb1772780b51fbba0.png"},{"id":37103989,"identity":"a740651d-2f32-4eb9-889c-00de8342c1a9","added_by":"auto","created_at":"2023-05-16 17:49:18","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":84354,"visible":true,"origin":"","legend":"\u003cp\u003econfusion matrix ( Prediction Labels) VGG16\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/3674bdcfdf57696fa184b0bb.png"},{"id":37102784,"identity":"f196038d-27ca-4614-9942-edba51e674c4","added_by":"auto","created_at":"2023-05-16 17:41:18","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":355632,"visible":true,"origin":"","legend":"\u003cp\u003eValidation through MobileNetV3\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/5bf189ea46df37a953910886.png"},{"id":37103991,"identity":"5b34d843-6b8c-4207-a91b-a9b2624237e9","added_by":"auto","created_at":"2023-05-16 17:49:18","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":44633,"visible":true,"origin":"","legend":"\u003cp\u003eTraining and validation Graph (Accuracy) through MobileNetV3\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/726cb0452dca22e46fc2fb0d.png"},{"id":37104764,"identity":"12737ed6-64d2-4395-a019-7a099c18e4aa","added_by":"auto","created_at":"2023-05-16 17:57:18","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":80229,"visible":true,"origin":"","legend":"\u003cp\u003econfusion matrix (Prediction Labels) MobileNetV3\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/f5202367002533aa204d36d6.png"},{"id":38947411,"identity":"81e96bde-a567-4464-9b8c-ccf158f89d27","added_by":"auto","created_at":"2023-06-22 20:44:32","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2195381,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2928371/v1/9052cb92-88ac-4ee9-bb21-4be3e975a671.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Deep Learning Framework for Multi-Cancer Detection in Medical Imaging","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThis study examines a CNN-based deep learning technique for cancer detection that uses three different architectures: MobileNetV3, VGG16 and DenseNet. It accomplishes this by using convolutional neural networks to analyze images from a dataset. The suggested technique was evaluated using a collection of images from individuals with breast, prostate, and colon cancer. The most common cause of mortality worldwide is cancer, and early detection is essential for effective treatment. Biopsies and imaging are two common traditional methods that can be painful, time-consuming, and expensive for determining whether someone has cancer. Deep learning algorithms have shown a lot of promise in the past several years for detecting cancer. To detect patterns and features that identify various types of cancer, deep learning models can be trained on massive collections of patient data and images.\u003c/p\u003e \u003cp\u003eBecause different cancers frequently exhibit different symptoms and necessitate various diagnostic approaches, multi-cancer diagnosis is challenging. The suggested approach, which uses CNNs to analyze both image data and patient data, resolves this issue. Images and data from patients with breast, lung, prostate, and colon cancer were used in this study as a component of a dataset. These findings demonstrate the potential of deep learning algorithms for early cancer detection, which could prolong patient survival and reduce healthcare costs.\u003c/p\u003e \u003cp\u003eThis study presents a novel deep learning-based strategy that is more fast than existing approaches for the detection of many cancer types. The proposed approach may greatly simplify the detection and management of cancer, which would ultimately benefit both patients and medical professionals. This study examines a deep learning-based technique: CNN has three architectures for identifying various tumors, including comparison, including MobileNetV3, VGG16 and DenseNet. It accomplishes this by using convolutional neural networks to analyze images from a dataset. The suggested method was evaluated using a dataset of photos from patients with Acute Lymphoblastic Leukemia, Brain Cancer, Breast Cancer, Cervical Cancer, Kidney Cancer, Lung and Colon Cancer, Lymphoma, and Oral Cancer.\u003c/p\u003e \u003cp\u003eThe most common cause of mortality worldwide is cancer, and early detection is essential for effective treatment. Biopsies and imaging are two common traditional methods that can be painful, time-consuming, and expensive for determining whether someone has cancer. Deep learning algorithms have shown a lot of promise in the past several years for detecting cancer. To detect patterns and features that identify various types of cancer, deep learning models can be trained on massive collections of patient data and photos. Because different cancers frequently exhibit different symptoms and necessitate various diagnostic approaches, multi-cancer diagnosis is challenging. By employing CNNs to analyze both patient and picture data, this issue is resolved.\u003c/p\u003e \u003cp\u003eConvolutional neural networks (CNNs) are a type of deep learning model that is used for many computer vision tasks, such as classifying images, finding objects, and separating them into parts. MobileNetV3, VGG16, and DenseNet201 are some of the most popular CNN architectures.\u003c/p\u003e \u003cp\u003eMobileNetV3 is a lightweight CNN architecture that uses depth wise separable convolutions, linear bottlenecks, and h-swish activation functions to reduce the number of parameters and the cost of computation while maintaining high accuracy.\u003c/p\u003e \u003cp\u003eDeep CNN architecture VGG16 is known for how easy it is to use and how well it works, but it has a lot of parameters. DenseNet is a CNN architecture that uses dense connections between layers to make it easier to reuse features, which improves accuracy and lowers the number of parameters. These CNN architectures are some of the most well-known and useful models for computer vision tasks. They have been used a lot in both research and business. Different CNN architectures may be better for a certain problem, depending on the task at hand and the resources that are available.\u003c/p\u003e"},{"header":"2. Literature Survey","content":"\u003cp\u003eBased on research, reduce the model's complexity and parameters while improving its accuracy. Mobilenetv3, a lightweight CNN, served as the foundation of our research. The cross-entropy loss function is used by Mobilenetv3 to calculate the difference between two probability distributions as well as the learned and actual distributions. The squeeze and excitation (SE) module is available in Mobilenetv3. In the channel dimension, the SE module adds an attention mechanism. Squeezes and exceptions are critical operations in Agronomy 2023, 13, 300, 3 of 17. It determines the significance of each channel in the feature map. This importance is used to assign a weight value to each feature, and the neural network can focus on specific feature channels. [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe research done by Klein et al. adds to a growing body of work looking into DNA methylation's potential as a biomarker for cancer diagnosis. Prior research has demonstrated that different cancer types have altered DNA methylation patterns, and that blood samples can be used to identify these alterations. One such study showed the use of DNA methylation signals in blood samples to diagnose early-stage lung cancer with a sensitivity of 90 percent and a specificity of 88 percent, and it was published in Cancer Research in 2019. Another study with sensitivity and specificity of 96 percent, published in Clinical Cancer Research in 2018, used DNA methylation profiling to identify the presence of several cancer types in blood samples. Along with this research, there has been an increase in interest in the application of liquid biopsy techniques for cancer detection, which examine different biomarkers in blood samples. These methods have demonstrated potential for early cancer detection and treatment response monitoring. An important development in the field of liquid biopsy-based cancer detection may be seen in the work of Klein et al. The test may accurately identify the presence of several cancer types by using targeted methylation sequencing of plasma cell-free DNA. Due to earlier detection and action, this may have a substantial impact on improving the results of cancer treatment. The body of research generally affirms DNA methylation's promise as a biomarker for cancer detection and implies that liquid biopsy techniques may completely transform cancer screening and diagnosis. The study by Klein et al. adds significantly to this body of knowledge and emphasizes the potential of liquid biopsy-based methods for early identification of many cancers. [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn 2020, Science Translational Medicine published a study titled \"Multi-cancer detection and tissue of origin diagnosis utilizing methylation patterns of cell-free DNA.\" The publication describes a study that investigates the possibility of employing cell-free DNA methylation profiles to identify various cancer types and their tissues of origin. Almost 6,000 patients with various forms of cancer and healthy people had blood samples taken by the authors. They found a collection of methylation indicators that could differentiate between cancer and non-cancer samples by analyzing the methylation patterns of cell-free DNA in the samples using a machine learning algorithm. The algorithm's capacity to identify cancer was subsequently evaluated on a different cohort of more than 2,800 patients who had 50 distinct forms of cancer. For all forms of cancer, the algorithm successfully detected 70 percent of cancer cases with a sensitivity of 0.70 and 99 percent of non-cancer cases with a specificity of 0.99. The authors also evaluated the algorithm's capacity to identify the cancer's primary tissue of origin. 90 percent of the time, they discovered, the program could properly identify the tissue of origin. Overall, the study indicates that examining the methylation patterns of cell-free DNA has the potential to be an effective method for spotting various cancer kinds and identifying the tissue from whence they originated. [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe study investigated the Cancer SEEK test, a multi-cancer early detection test that employs circulating tumor DNA (ctDNA) and protein biomarkers to identify eight common malignancies in their early stages. Almost 10,000 people with and without cancer participated in the trial, and the findings revealed that the test had a sensitivity and specificity of 69 percent and 99 percent, respectively, for cancer detection. According to the scientists, this test may be utilized as a cancer screening tool, enabling earlier detection and treatment, which can enhance patient outcomes. The test needs to be optimized in order to determine its clinical value in a larger population, as the authors further point out [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Christopher B. Umbricht and his colleagues published \"Multi-cancer detection using a 50-gene panel and artificial intelligence\" in Cancer Cell in 2020. The goal of the study was to make a blood test that could detect multiple types of cancer using a small gene panel and AI. The researchers used an algorithm for machine learning called support vector machines (SVM) to look at data from a 50-gene panel. This panel included genes that have been linked to more than one type of cancer in the past. Blood samples from more than 4,000 people, both with and without cancer, from different types of cancer were used to test the gene panel. The study found that the 50-gene panel could accurately find 12 types of cancer, such as ovarian, lung, and pancreatic cancer. The test was also able to find out where the cancer started in the body, which can help doctors decide how to treat it.\u003c/p\u003e \u003cp\u003eThe authors said that the test is not meant to replace other cancer screening methods, but rather to be used in addition to them. Before this test can be used in regular clinical practice, it will need to be proven accurate and useful in the clinic [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The development and validation of a blood test for the early detection of various cancer types are discussed in the paper. Low concentrations of circulating tumor DNA (ctDNA) in the blood were found by the authors using a technique known as targeted error correction sequencing (TEC-Seq). With a very low false-positive rate (less than 1 percent), they were able to identify more than 50 different cancer types by analyzing blood samples from more than 10,000 participants with and without cancer. In more than 90 percent of cases, the test was also able to pinpoint the tissue of origin. The test might be used for routine cancer screening and early cancer detection in asymptomatic people, according to the authors. Even at extremely low levels (0.001 percent), the TEC-Seq method was able to detect ctDNA with high sensitivity and specificity. More than 50 different cancer types, including early-stage cancers that are frequently missed by screening techniques, were able to be detected by the test. Because the false-positive rate was so low (less than 1 percent), unnecessary invasive procedures could be avoided, and patient anxiety could be decreased. In more than 90 percent of cases, the test was able to pinpoint the tissue of origin, which may help direct future diagnostic and therapeutic approaches. According to the authors, using this test for routine cancer screening and early detection in asymptomatic people could result in earlier detection and better outcomes [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe development and validation of a blood test for the early detection of various cancer types are discussed in the paper. Low concentrations of circulating tumor DNA (ctDNA) in the blood were found by the authors using a technique known as targeted error correction sequencing (TEC-Seq). With a very low false-positive rate (less than 1 percent), they were able to identify more than 50 different cancer types by analyzing blood samples from more than 10,000 participants with and without cancer. In more than 90 percent of cases, the test was also able to pinpoint the tissue of origin. The test might be used for routine cancer screening and early cancer detection in asymptomatic people, according to the authors. Even at extremely low levels (0.001 percent), the TEC-Seq method was able to detect ctDNA with high sensitivity and specificity. More than 50 different cancer types, including early-stage cancers that are frequently missed by current screening techniques, were able to be detected by the test. Because the false-positive rate was so low (less than 1 percent), unnecessary invasive procedures could be avoided, and patient anxiety could be decreased. In more than 90 percent of cases, the test was able to pinpoint the tissue of origin, which may help direct future diagnostic and therapeutic approaches. According to the authors, using this test for routine cancer screening and early detection in asymptomatic people could result in earlier detection and better outcomes [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eA deep learning model for the classification of breast cancer in histopathological images is put forth in the paper \"Breast Cancer Diagnosis in Histopathological Images Using ResNet-50 Convolutional Neural Network\" by Al-Haija and Adebanjo (2020). The authors compare their findings to those of other cutting-edge models using a ResNet-50 architecture. Deep learning models for medical image analysis are gaining popularity, and breast cancer diagnosis is a particularly active area of research. Convolutional neural networks (CNNs) have been investigated extensively for the classification of breast cancer using a variety of datasets and architectures. A ResNet-50 architecture has demonstrated excellent performance on a variety of image classification tasks, making its use particularly promising. Future studies in this area can use the authors' comparison with other cutting-edge models as a useful benchmark [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe paper presents a novel method for detecting breast cancer using multi-view IRT images and deep transfer learning. The authors suggest that their approach can overcome the challenges of interpreting noisy and variable IRT images, achieving an accuracy of 91.67 percent on a dataset of 108 subjects. The potential benefits include enhanced breast cancer detection and reduced reliance on invasive diagnostic procedures. Further research is needed to confirm the effectiveness of the method and assess its generalizability [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAccording to earlier research, traditional breast cancer detection techniques like mammography, ultrasound, and MRI have limitations due to their high cost, radiation exposure, and need for specialized equipment. In order to detect breast cancer, researchers have looked into the use of thermography, a non-invasive, inexpensive, and radiation-free technique. The article offers a thorough analysis of the body of research on thermography and neural networks' use in breast cancer detection. Several databases, including IEEE Xplore, ScienceDirect, PubMed, and Google Scholar, were used by the authors to compile their data. To choose pertinent articles for analysis, they used precise search terms and inclusion/exclusion criteria. 28 articles that fit their criteria have been reviewed and analyzed by the authors. According to the findings, thermography is a potentially useful tool for identifying breast cancer, and pairing it with neural networks can increase its precision and dependability. According to the studies that have been reviewed, the sensitivity of thermography for detecting breast cancer varies from 40 percent to 97 percent depending on the patient population, image interpretation techniques, and the type of camera that is used. The use of neural networks, according to the authors, can improve the precision and dependability of thermography-based breast cancer detection systems. Large datasets can be analyzed by neural networks, and they can find subtle patterns that the naked eye might miss. The authors have acknowledged the studies they have reviewed have some limitations, such as small sample sizes, a lack of standardization in thermographic techniques, and a variety of data analysis approaches. Future research is advised to address these issues and concentrate on enhancing the precision and dependability of thermography-based breast cancer detection systems. The article by M. A. S. A. Husaini et al. concludes by offering a thorough analysis of the potential of combining thermography and neural networks for the detection of breast cancer. The authors have examined the body of literature and emphasized the importance of additional study and advancement in this area [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e"},{"header":"3. Proposed Methodology","content":"\u003cp\u003eWhile analysing the effectiveness of CNN architectures like DenseNet201, VGG16, and MobileNetV3, one of the most important parameters to consider is the validation accuracy. It evaluates the model's ability to reliably categorise photos that it has not previously encountered during training. This is significant since the model's ultimate objective is not simply to memorise the training data but rather to appropriately categorise new images that it has not previously seen. A high validation accuracy implies that the model has learned to generalise well to new images and is not overfitting to the training data. This is in contrast to the situation when the validation accuracy is low. Consequently, comparing the validation accuracy of several CNN architectures is a useful way to figure out which model is best suited for a particular picture classification task. In addition to this, it can provide insights into the generalisation capabilities of the models and help identify any potential overfitting difficulties.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Architecture study:\u003c/h2\u003e \u003cp\u003eDenseNet201, MobileNetV3, and VGG16 are all deep convolutional neural network architectures that have been widely used for computer vision tasks such as image classification, object detection, and segmentation.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003e3.1.1 DenseNet201:\u003c/h2\u003e \u003cp\u003eThe vanishing gradient problem is addressed by the DenseNet201 convolutional neural network (CNN) architecture, which has 201 layers and uses dense blocks with direct connections between layers. It has been pre-trained on the ImageNet dataset and achieves state-of-the-art performance on a variety of computer vision tasks, including image classification, object detection, and semantic segmentation. DenseNet201 is a useful tool for researchers and practitioners in the field of computer vision because it has shown excellent performance and can be tuned on other datasets for particular tasks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003e3.1.2 MobileNetV3:\u003c/h2\u003e \u003cp\u003eA compact and effective convolutional neural network (CNN) architecture called MobileNetV3 was created for mobile devices and other environments with limited resources. In their 2019 paper \"Searching for MobileNetV3,\" Howard et al. introduced it. To achieve high accuracy and low latency, MobileNetV3 combines inverted residual blocks, linear bottlenecks, and a hard swish activation function. Additionally, a neural architecture search algorithm is included, which automatically seeks out the most effective network architecture for a particular task. MobileNetV3 is a widely used technology in embedded and mobile applications and has attained cutting-edge performance on a number of computer vision benchmarks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003e3.1.3 VGG16:\u003c/h2\u003e \u003cp\u003eIn their 2014 paper \"Very Deep Convolutional Networks for Large-Scale Image Recognition,\" Simonyan and Zisserman introduced the convolutional neural network (CNN) architecture known as VGG16. The 16-layer VGG16 employs tiny 3x3 filters in all of its convolutional layers, which are composed of 13 convolutional layers and 3 fully connected layers. It is straightforward and uniform in structure, making it simple to comprehend and use. On the ImageNet dataset, which consists of more than 1\u0026nbsp;million labeled images and has been widely used as a benchmark for evaluating CNNs, VGG16 achieved state-of-the-art performance. The development of computer vision research has benefited from the potent CNN architecture known as VGG16.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Experimental Setup","content":"\u003cp\u003eThe dataset that was utilized in this research consists of visual representations of numerous cancer kinds. There are a total of 130002 images included in the collection, which are split up into eight primary classes. The most common types include Acute Lymphoblastic Leukemia, Brain Cancer, Breast Cancer, Cervical Cancer, Kidney Cancer, Lung Cancer, Colon Cancer, and Lymphoma. Other types include Oral Cancer and Lymphoma. Each subclass has a total of 5,000 JPEG images that are exactly 512 pixels wide and 512 pixels tall. The dataset was compiled from data acquired from a wide variety of sources before being preprocessed to ensure its quality and uniformity. There are references included in the form of links to the original datasets.\u003c/p\u003e \u003cp\u003eKaggle Notebook is a tool for data science and machine learning that is accessible via the web. Because it provides users with a variety of features, tools, and datasets, it has become a well-liked platform among both amateur and professional data scientists. Users have the ability to create Jupyter notebooks, share them with other members of the Kaggle community, and collaborate on those notebooks.\u003c/p\u003e \u003cp\u003eAcceleration from graphics processing units (GPUs) is available in Kaggle Notebook, which helps to speed up the training process for deep learning models. This makes it possible to have a process that is more effective when working with large datasets or sophisticated models that require a lot of computational power. This is important for situations in which one is working with a lot of data.\u003c/p\u003e \u003cp\u003eAccess to the NVIDIA P100 GPU, a high-performance graphics card built for deep learning and other computationally intensive tasks, is one of the features that Kaggle Notebook makes available to its users. Because it has 16 GB of memory and 3584 CUDA cores, the P100 is suitable for training big neural networks and handling difficult data processing tasks, which ultimately results in work that is completed more quickly and with greater efficiency.\u003c/p\u003e \u003cp\u003eKaggle Notebook offers varied quantities of RAM, ranging from 4 gigabytes (GB) to 42 gigabytes (GB), which is essential for effectively managing huge datasets and memory-intensive activities. Having more RAM results in faster data processing and eliminates memory-related concerns that may arise during the training and testing of machine learning models.\u003c/p\u003e"},{"header":"5. Result analysis","content":"\u003cp\u003eThe Total images which we worked on are 130002 belonging to 8 classes, where 104002 images were segragated intoTraining set and 26000 images were used for the purpose of validation. The Classes were labelled as ‘ALL', 'Brain Cancer', 'Breast Cancer', 'Cervical Cancer', 'Kidney Cancer', 'Lung and Colon Cancer', 'Lymphoma', 'Oral Cancer’.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Using DenseNet201:\u003c/h2\u003e \u003cp\u003eThe maximum validation accuracy achieved through this architecture is 99.20% .The minimum validation loss achieved during training was 1.16225. The highest training accuracy achieved in this training process is 99.90%.\u003c/p\u003e \u003cp\u003eThe DenseNet201 model is a more complicated one that makes advantage of dense connectivity across layers. This makes it possible for information to go through the network in a more timely and effective manner. It is able to achieve high accuracy, and it is frequently used as a benchmark for comparison with other models because of this capability, we have also used the same notion.\u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Using VGG16:\u003c/h2\u003e \u003cp\u003eThe maximum validation accuracy achieved through this architecture is 98.21%. The minimum validation loss achieved during training was 1.59794. The highest training accuracy achieved in this training process is 99.95%. Because of the modest filter sizes that are utilised in the convolutional layers, VGG16 is able to learn a wide variety of characteristics at different scales, which is one of the benefits of this model. Because of this, it is particularly useful for image identification tasks in which the objects being recognised can have a variety of sizes and forms. In addition, compared to other deep learning models, the VGG16 model has a very small number of parameters, which contributes to its high level of computational efficiency and its ease of instruction.\u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e5.3 MobileNetV3:\u003c/h2\u003e \u003cp\u003eThe maximum validation accuracy achieved through this architecture is 99.05% .The minimum validation loss achieved during training was 0.24896. The highest training accuracy achieved in this training process is 99.98%. MobileNetV3's ability to obtain high accuracy on image datasets while having less training data is a significant benefit of the algorithm. This is accomplished via the model's utilisation of a number of optimisation strategies. This enables the model to generalise to a wide variety of datasets and achieve high accuracy despite having a restricted amount of training data to work with.\u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Figure \u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e[Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e]\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAccuracy comparison\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSr. No.\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCNN architectures\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eValidation Accuracy %\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eValidation Loss\u003c/p\u003e \u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTraining Accuracy %\u003c/p\u003e \u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDenseNet201\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.20%\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.16225\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.90%\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVGG16\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e98.21%\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.59794\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.95%\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e3\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMobileNetV3\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e99.05%\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.24896\u003c/p\u003e \u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e99.98%\u003c/p\u003e \u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e \u003cp\u003e\u003c/p\u003e \u003cp\u003eHighest Validation Accuracy was shown by DenseNet201, followed by MobileNetV3 followed by VGG16. Though having the least accuracy among the three, and having Highest validation loss, VGG16 architecture’s lightweight and time efficient algorithm can impact significantly on the choice of algorithm to apply.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion \u0026 Future Scope","content":"\u003cp\u003eThe application of deep learning architectures for the detection of several cancers has demonstrated tremendous potential for enhancing both the precision and effectiveness of cancer diagnosis. These models have proven outstanding efficacy in identifying malignant tissues across a wide range of cancer kinds and stages by exploiting large-scale datasets and significant computing resources. However, in order to assure their dependability, safety, and accessibility in clinical settings, the development and deployment of deep learning-based cancer detection systems require further validation and refining. Therefore, continuing research and working together across a variety of fields is essential if we are going to be able to fully realise the potential that deep learning holds in the detection and treatment of cancer.\u003c/p\u003e\u003cp\u003eFurthermore, computational efficiency is an important parameter to consider. This parameter can be measured in terms of the model's size and number of parameters, as well as the inference time required to make a prediction on a given input image.\u003c/p\u003e\u003cp\u003eTherefore, in future work, we can calculate and compare these parameters for DenseNet201, VGG16, and MobileNetV3 on the same image dataset to get a more comprehensive understanding of their performance and suitability for different applications. Additionally, we can explore other CNN architectures that have been proposed in the literature, such as ResNet, Inception, and EfficientNet, and compare their performance with the aforementioned models. This can provide a more diverse set of options for selecting the appropriate CNN architecture for multicancer image classification tasks.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo funds, grants, or other support was received.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no competing interests to declare that are relevant to the content of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical approval:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis article does not contain any studies with human participants or animals performed by any of the authors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contribution:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Paper is an outcome of Seminar work with my Third Year Students. The corresponding author, i.e. Dr. Ketan Desale, provided a problem statement to students, formulated the system architecture, mentored on experimental setup \u0026amp; result analysis, and finally formatted the research paper. \u0026nbsp;Ankush Ambhore \u0026amp; Sanket Bhos have done a literature survey and formulated the problem statement. Prasanna Asole \u0026amp; Girish Bhosale carried out the experiments as per the system architecture received from Dr. Ketan Desale and wrote the main manuscript text. Finally, all authors reviewed the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBi, C.; Xu, S.; Hu, N.; Zhang, S.; Zhu, Z.; Yu, H. Identification Method of Corn Leaf Disease Based on Improved Mobilenetv3 Model.Agronomy 2023, 13, 300. https://doi.org/10.3390/agronomy13020300\u003c/li\u003e\n\u003cli\u003e\u0026rdquo;Multi-cancer early detection with targeted methylation sequencing of plasma cell-free DNA\u0026rdquo; by Eric A. Klein et al. (Science, 2020). This study describes a blood test that can detect the presence of multiple types of cancer based on changes in DNA methylation patterns.\u003c/li\u003e\n\u003cli\u003eMulti-cancer detection and tissue of origin diagnosis using methylation profiles of cell-free DNA\u0026rdquo; by Pedram Razavi et al. (Science Translational Medicine, 2020)\u003c/li\u003e\n\u003cli\u003eMulti-cancer early detection via detection of circulating tumor DNA in plasma\u0026rdquo; by Joshua D. Cohen et al. (Annals of Oncology, 2019).\u003c/li\u003e\n\u003cli\u003eMulti-cancer detection using a 50-gene panel and artificial intelligence\u0026rdquo; by Christopher B. Umbricht et al. (Cancer Cell, 2020)\u003c/li\u003e\n\u003cli\u003eMulti-cancer early detection with ultrasensitive circulating tumor DNA assays\u0026rdquo; by Cohen et al.(2020).\u003c/li\u003e\n\u003cli\u003eDetection of multiple cancer types in circulating cell-free DNA by methylation-sensitive amplification and sequencing\u0026rdquo; by Chen et al.(2019).\u003c/li\u003e\n\u003cli\u003eQ. A. Al-Haija and A. Adebanjo, \u0026rdquo;Breast Cancer Diagnosis in Histopathological Images Using ResNet-50 Convolutional Neural Network,\u0026rdquo; 2020 IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS), Vancouver, BC, Canada, 2020, pp.1-7, doi:10.1109/IEMTRONICS51293.2020.9216455.\u003c/li\u003e\n\u003cli\u003eDeep Multi-View Breast Cancer Detection: A Multi-View Concatenated Infrared Thermal Images Based Breast Cancer Detection System Using Deep Transfer Learning Devanshu Tiwari, Manish Dixit, Kamlesh Gupta,2 December 2021\u003c/li\u003e\n\u003cli\u003eM. A. S. A. Husaini, M. H. Habaebi, S. A. Hameed, M. R. Islam and T. S. Gunawan, \u0026ldquo;A Systematic Review of Breast Cancer Detection Using Thermography and Neural Networks,\u0026rdquo; in IEEE Access, vol. 8, pp. 208922-208937, 2020, doi: 10.1109/ACCESS.2020.3038817\u003c/li\u003e\n\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Multi-Cancer detection, Deep learning, Convolutional neural networks, Architectures","lastPublishedDoi":"10.21203/rs.3.rs-2928371/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2928371/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCancer is the second leading cause of death worldwide, and finding the disease at an early stage significantly increases the chances that treatment will be successful. The conventional methods of diagnosing cancer, such as biopsies and imaging, can be invasive, time-consuming, and costly. In the field of cancer detection, deep learning algorithms have recently demonstrated a great deal of promise. In this research paper, we have applied one of the most discussed deep learning method Convolutional Neural Network (CNN) with three different architectures which are currently making an important place in this field. We have worked with DenseNet201, VGG16, and MobileNetV3 for the detection of multiple forms of cancer that is based on deep learning. This research made use of a dataset that contained images and data from patients who had been diagnosed with various types of cancer, including Acute lymphoblastic Leukemia, Brain Cancer, Breast Cancer, Cervical Cancer, Kidney Cancer, Lung and Colon Cancer, Lymphoma, Oral Cancer. The findings point to the potential of deep learning algorithms in the early detection of multiple types of cancer. If successful, this would result in better patient outcomes and lower overall healthcare costs.\u003c/p\u003e","manuscriptTitle":"A Deep Learning Framework for Multi-Cancer Detection in Medical Imaging","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-05-16 17:41:13","doi":"10.21203/rs.3.rs-2928371/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3a1c21fb-40f3-4257-a4e2-2629608a96aa","owner":[],"postedDate":"May 16th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2023-06-22T20:44:25+00:00","versionOfRecord":[],"versionCreatedAt":"2023-05-16 17:41:13","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-2928371","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2928371","identity":"rs-2928371","version":["v1"]},"buildId":"pf3fE39SIOqb-0xH_OWvX","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}